Gauge Freedom and the Coulomb Gauge

The previous article introduced a field that earns its keep and then admitted, in fine print, that the field is not unique: add any gradient to $\vec A$ and every magnetic fact stays exactly the same. This article takes that fine print and makes it the whole page — because the redundancy of $\vec A$ is not a defect to be apologized for but a resource to be spent, and learning to spend it is a skill physics reuses from magnet design to the Standard Model. The honest questions come in a fixed order. How much freedom is there? Exactly the gradients — no more, no less: any two potentials for the same field differ by $\nabla\lambda$ , a theorem this article proves rather than asserts. What does the freedom mean? That $\vec A$ is a description with more coordinates than the physics has content, like altitudes quoted from an arbitrary sea level — except that for a vector field the "sea level" is a whole scalar function's worth of choice, and the gauge-invariant content (the field $\vec B$ , the fluxes $\oint\vec A\cdot\mathrm{d}\vec\ell$ ) is what every observer must agree on. And how do you spend it? Impose a condition that picks a convenient representative. Magnetostatics' standard pick is the Coulomb gauge, $\nabla\cdot\vec A=0$ — always reachable, essentially unique once boundary behavior is fixed, and profitable: with it, Ampère's law collapses into three copies of Poisson's equation, and the entire electrostatic solving apparatus of chapters E1–E2 transfers to magnetism wholesale. To keep the freedom from staying abstract, the article displays it concretely: one uniform magnetic field, dressed in visibly different potentials — swirling, striped, tilted — all provably the same physics. If you can watch those maps disagree while the territory stays fixed, you own the concept the twentieth century built its physics on.

Key takeaways

Gauge freedom: $\vec A\to\vec A+\nabla\lambda$ leaves $\vec B$ unchanged — and that is all the freedom there is: two potentials with equal curls differ by a gradient (proved here from the E0 converse theorem).
Gauge-invariant = physical: $\vec B$ , fluxes $\oint\vec A\cdot\mathrm{d}\vec\ell$ , and everything measurable. Gauge-dependent = descriptive: pointwise $\vec A$ , the canonical momentum's split into $m\vec v$ and $q\vec A$ .
The Coulomb gauge $\nabla\cdot\vec A=0$ is always reachable: repairing a candidate potential means solving one Poisson equation $\nabla^2\lambda=-\nabla\cdot\vec A_{\rm old}$ , which E1–E2 machinery guarantees.
Residual freedom after gauge fixing = harmonic gauge functions ( $\nabla^2\lambda=0$ ); demanding $\vec A\to0$ at infinity kills them, pinning the representative.
The payoff: in Coulomb gauge, $\nabla\times\vec B=\mu_0\vec J$ becomes $\nabla^2\vec A=-\mu_0\vec J$ — three independent Poisson equations, solved by the previous article's integral.
One uniform field, many uniforms: $-By\,\hat x$ , $+Bx\,\hat y$ , and $\tfrac12\vec B\times\vec r$ all describe $B\hat z$ ; all three are even Coulomb — differing by harmonic gauge functions.
KEY IDEA: only $\nabla\times\vec A$ is physics; the rest of $\vec A$ is choice — and choosing $\nabla\cdot\vec A=0$ converts magnetostatics into electrostatics-shaped Poisson problems.

The core insight

Only $\vec B=\nabla\times\vec A$ is physical; $\vec A$ is fixed only up to $\vec A\to\vec A+\nabla\lambda$ . Imposing $\nabla\cdot\vec A=0$ spends this freedom and turns Ampère's law into $\nabla^2\vec A=-\mu_0\vec J$ .

Where this lives today

Weyl's word gauge (Eichung — calibration, as of a measuring rod) entered physics in 1918 through a failed theory and stayed through a triumphant one. The failure: Weyl proposed that the scale of length might vary point to point, its arbitrariness compensated by a field with $\vec A$ 's transformation law; Einstein's lethal objection — atomic spectral lines would then depend on a clock's history, which they observably do not — killed the idea in a footnote-length exchange. The rescue came with quantum mechanics: London and Fock noticed that the wavefunction's phase, not the meter stick, is the locally arbitrary quantity, and Weyl's 1929 relaunch — physics must be invariant under $\psi\to e^{iq\lambda/\hbar}\psi$ together with $\vec A\to\vec A+\nabla\lambda$ — turned this article's redundancy into a generating principle: demand the local symmetry, and the electromagnetic field, its coupling to charge, and charge conservation itself are forced into existence. That template, generalized by Yang and Mills (1954) from phase rotations to larger symmetry groups, is the architecture of the Standard Model: the strong and weak interactions are gauge freedoms with matrix-valued $\lambda$ 's, their force fields the corresponding $\vec A$ 's. On the workbench, meanwhile, gauge choices are daily engineering: magnetostatic finite-element solvers impose Coulomb-type conditions to make their matrix problems invertible (the numerical echo of "fix the representative or the equations don't determine it"), quantum chemists choose gauges that minimize basis-set artifacts in molecular magnetism, and lattice QCD spends supercomputer-years managing exactly the redundancy this article taught you to spend with a pencil.

Origins · 1 minHermann Weyl and how this was discovered

Hermann Weyl

1885–1955 · Zürich, Göttingen, Princeton

Weyl was the rare mathematician whom physicists read for physics — successor to Hilbert's tradition at Göttingen, author of the first great book on general relativity (Raum, Zeit, Materie, 1918), and the man who named the gauge principle before physics was ready for it. His 1918 unification made length itself path-dependent, with an "Eich" (calibration) field absorbing the arbitrariness; Einstein admired the mathematics and demolished the physics in the same letter — atoms are universal clocks, and they disagree. Weyl took the defeat and banked the structure. When quantum mechanics revealed the wavefunction's phase as nature's genuinely arbitrary local quantity, he reissued the principle in its modern form (1929): local phase freedom, compensated by the electromagnetic potential, with charge conservation as the Noether dividend. He fled Göttingen in 1933 for Princeton, where his group-theoretic methods seeded half of modern particle physics; "my work always tried to unite the true with the beautiful," he said, "but when I had to choose, I usually chose the beautiful" — and in the gauge principle, unusually, he never had to choose.

History

1845

— Franz Neumann and later workers compute with vector potentials, quietly aware that different choices give the same forces.

1873

— Maxwell's Treatise uses A centrally; Heaviside and Hertz later strip the potentials out as unphysical scaffolding.

1918

— Weyl proposes gauge (scale) invariance to unify gravity and electromagnetism; Einstein's clock objection sinks it.

1929

— Weyl, after London and Fock, relaunches gauge invariance as local phase symmetry — the modern principle, with charge conservation as its Noether charge.

1954

— Yang and Mills generalize the gauge principle to non-abelian groups, the architecture of the Standard Model.

1959

— Aharonov and Bohm show the potential's gauge-invariant loop integral is quantum-mechanically observable, closing the "mere scaffolding" era for good.

The physical picture

Start with the scalar case you already trust. The electrostatic potential $V$ was defined only up to a constant: sea-level Amsterdam versus sea-level Bombay, every altitude shifted, every altitude difference — the physics — untouched. Gauge freedom is that idea grown to vector size, and the growth is dramatic: instead of one number's worth of convention, $\vec A$ carries a full scalar function's worth. At every point you may shift $\vec A$ by the local value of $\nabla\lambda$ for any smooth $\lambda(\vec r)$ whatsoever, and no compass, no force measurement, no flux reading will ever notice. The right mental model is maps versus territory. The territory is $\vec B$ — and, if you want the complete inventory, the gauge-invariant functionals of $\vec A$ such as its closed-loop circulations (which the Aharonov–Bohm effect certifies as genuine territory too). A choice of gauge is a choice of map projection: Mercator distorts Greenland, a polar projection distorts the equator, and arguing about which map is "true" is a category error. But map choice is not idle — navigators pick the projection that makes their route a straight line. So does physics: the Coulomb gauge is the projection in which magnetostatics looks like electrostatics, the Lorenz gauge (kept on a named shelf until the radiation chapter) is the one in which relativity looks manifest, and the Landau gauge is the one in which a quantum charge in a uniform field separates variables. Here is one territory wearing three uniforms — the same constant field $B\hat z$ , dressed three ways:

Figure

One field, three gauges — swirling, striped along y, striped along x: identical physics in different coordinates of description

Look at what the three panels disagree about: where $\vec A$ vanishes (the center; the line $y=0$ ; the line $x=0$ ), which direction it points, whether it swirls or stripes. None of that is askable in the lab. Look at what they agree about: the curl ( $B\hat z$ everywhere), and the circulation around any closed loop you draw (trace one — a loop drawn about the swirl's centre gains circulation from every leg, though one drawn off-centre need not; the stripes gain on one leg what they don't lose on the return; both totals come out $B\times$ area). The agreement is the physics. And one more agreement, quieter but load-bearing for this article: all three have $\nabla\cdot\vec A=0$ . Even after imposing the Coulomb gauge, some freedom survives — these three differ by gauge functions that happen to be harmonic — which is exactly why the fixing story below has a second chapter about boundary conditions.

The freedom, measured exactly

First, the full extent of the ambiguity — both directions, because half the theorem is usually asserted on credit.

\vec A'=\vec A+\nabla\lambda \;\;\Longrightarrow\;\; \nabla\times\vec A'=\nabla\times\vec A \qquad (1)

where: $\lambda(\vec r)$ — any smooth scalar (the gauge function). This direction is the identity $\nabla\times(\nabla\lambda)\equiv0$ : gradients have no curl, so a gradient added to a potential is invisible to the field. The converse is the substantive claim: that gradients are the only invisible additions.

Derivation AAll of the freedom is a gradient

The goal is the converse theorem: if $\vec A_1$ and $\vec A_2$ generate the same magnetic field, $\nabla\times\vec A_1=\nabla\times\vec A_2$ , then their difference is the gradient of some scalar — so equation (1) is not merely a symmetry of the description but the entire symmetry, and "fix the gauge" means "choose $\lambda$ ," nothing more. The strategy is to hand the difference field to the E0 converse theorem and then be honest about the topological fine print, which the solenoid — of course — probes.

Step 1. Form the difference. Let $\vec W=\vec A_1-\vec A_2$ . By hypothesis

\nabla\times\vec W=\nabla\times\vec A_1-\nabla\times\vec A_2=\vec 0

everywhere in the region: $\vec W$ is curl-free.

Step 2. Invoke the converse theorem. The vector-calculus chapter proved (via the fundamental theorem for gradients and Stokes) that on a well-behaved domain, a curl-free field is a gradient: there exists $\lambda(\vec r)$ , constructed explicitly as the path integral $\lambda(\vec r)=\int_{\mathcal O}^{\vec r}\vec W\cdot\mathrm{d}\vec\ell$ from any fixed base point, path-independent precisely because $\nabla\times\vec W=0$ kills the loop integrals. Hence

\vec A_1=\vec A_2+\nabla\lambda .

Step 3. State the fine print. "Well-behaved" means simply connected: every loop shrinkable to a point. In a region with a hole — the exterior of a solenoid is the canonical example — a curl-free $\vec W$ can still have nonzero circulation around the unshrinkable loop, and then $\lambda$ comes back multivalued. Physics polices this automatically: two legitimate potentials for the same field must agree on every circulation ( $\oint\vec A\cdot\mathrm{d}\vec\ell=\Phi_B$ is gauge-invariant), so their difference $\vec W$ has zero circulation even around the hole, and a single-valued $\lambda$ survives. The multivalued case is not pathology but physics elsewhere: it is why no scalar potential can describe $\vec B$ itself around a current-carrying wire. Why the theorem earns its keep: it upgrades every "choose a gauge" instruction from hopeful to exhaustive. When we impose $\nabla\cdot\vec A=0$ below, we search only over gauge functions $\lambda$ — and this theorem is the license that nothing outside that search could have been missed. *As a concrete check: symmetric minus Landau gives $\tfrac12(-By,Bx,0)-(-By,0,0)=\tfrac12(By,Bx,0)=\nabla(\tfrac12Bxy)$ — a single-valued gauge function, exactly as the theorem demands.*

What, then, is physical? The gauge-proof inventory assembled so far: the field $\vec B$ itself; every closed-loop circulation $\oint\vec A\cdot\mathrm{d}\vec\ell$ (shifting by $\oint\nabla\lambda\cdot\mathrm{d}\vec\ell=0$ ); hence every flux, every Aharonov–Bohm phase, every EMF the induction chapter will build. What is not: the value of $\vec A$ at a point; the direction of $\vec A$ ; the location of its zeros; and — a trap worth naming early — the split of canonical momentum $\vec p=m\vec v+q\vec A$ into its two parts. Any calculation whose final answer changes under $\vec A\to\vec A+\nabla\lambda$ has a bug by definition; this "gauge-invariance audit" is the cheapest error detector in field theory.

Spending the freedom: the Coulomb gauge

A gauge condition is a constraint imposed to nail down the representative. The magnetostatic standard:

\nabla\cdot\vec A=0 \qquad\text{(Coulomb gauge)} . \qquad (2)

where: the name honors the payoff — in this gauge the potential obeys Coulomb-style Poisson equations. Two facts make (2) a legitimate choice rather than an assumption about the world: it can always be reached, and what remains afterward is harmless. Both are one derivation — and the whole logical pipeline fits in one picture:

Figure

The two-lock door of gauge fixing — the differential condition narrows the family, boundary data picks the point

Go deeper · Derivation B · 2 minReaching the Coulomb gauge, and the freedom left over

Derivation BReaching the Coulomb gauge, and the freedom left over

The goal is twofold: show that from any starting potential $\vec A_0$ a gauge function $\lambda$ exists making $\vec A=\vec A_0+\nabla\lambda$ divergence-free; and characterize the residual freedom — the gauge transformations that preserve the condition once won. The strategy converts both questions into Poisson and Laplace problems respectively, letting the E1–E2 existence and uniqueness machinery do all the heavy lifting by name.

Step 1. Pose the repair as an equation for λ. Compute the divergence of the transformed potential:

\nabla\cdot(\vec A_0+\nabla\lambda)=\nabla\cdot\vec A_0+\nabla^2\lambda .

Demanding zero is the Poisson equation

\nabla^2\lambda=-\,\nabla\cdot\vec A_0 ,

with the known function $\nabla\cdot\vec A_0$ as its "source."

Step 2. Existence, on credit already banked. This is structurally the electrostatic problem $\nabla^2 V=-\rho/\varepsilon_0$ with $\rho/\varepsilon_0\to\nabla\cdot\vec A_0$ : for a localized source the explicit solution is the Coulomb-style integral $\lambda(\vec r)=\frac{1}{4\pi}\int\frac{\nabla'\cdot\vec A_0}{\mathfrak{r}}\mathrm{d}\tau'$ , and the E2 boundary-value machinery extends existence to bounded regions with reasonable boundary data. The Coulomb gauge is always reachable — not obvious a priori, but guaranteed by potential theory.

Step 3. The residual freedom is harmonic. Suppose $\vec A$ already satisfies (2). A further transformation $\vec A\to\vec A+\nabla\mu$ preserves it iff

\nabla\cdot(\vec A+\nabla\mu)=\nabla^2\mu=0 ,

i.e. iff $\mu$ is a harmonic function. The condition (2) does not pick one potential; it picks a family related by harmonic gauge functions — exactly how the three menagerie potentials, all Coulomb, coexist: they differ by $\tfrac12Bxy$ and its relatives, and $\nabla^2(xy)=0$ .

Step 4. Boundary conditions finish the job. For localized currents, add the demand $\vec A\to0$ at infinity. The residual $\nabla\mu$ must then also vanish at infinity with $\mu$ harmonic everywhere — and the E2 uniqueness arguments (mean-value/Liouville flavor: a harmonic function bounded with vanishing gradient at infinity has constant gradient, here zero) leave $\mu=$ const, $\nabla\mu=0$ . Coulomb gauge + decay at infinity = a unique $\vec A$ , and it is precisely the representative the previous article's integral formula computes. (The uniform-field menagerie evades this only because a uniform field filling all space never decays — an idealization with no infinity to anchor to.) Why this division of labor is elegant: the theory of gauge fixing contains no new mathematics — it is electrostatics' Poisson problem (Step 2) and Laplace uniqueness theory (Steps 3–4) rehired for a new client. Every hard theorem was proved two chapters ago. *The repair procedure in action: for $\vec A_0$ with $\nabla\cdot\vec A_0=2y$ (Worked example 2 below), Step 1 demands $\nabla^2\lambda=-2y$ , solved by $\lambda=-y^3/3$ ; the repaired potential is divergence-free and curl-unchanged.*

Now the purchase this was all for. Insert $\vec B=\nabla\times\vec A$ into Ampère's law and expand with the E0 double-curl identity:

\nabla\times\vec B=\nabla\times(\nabla\times\vec A)=\nabla(\nabla\cdot\vec A)-\nabla^2\vec A=\mu_0\vec J . \qquad (3)

where: $\nabla^2\vec A$ — the vector Laplacian, acting componentwise in Cartesian coordinates only (in cylindrical or spherical components the identity is the definition of the vector Laplacian, and componentwise Poisson fails — a classic trap). Equation (3) holds in every gauge, and in a general gauge it is a coupled, cluttered thing: the $\nabla(\nabla\cdot\vec A)$ term ties the three components together. The Coulomb gauge is precisely the choice that deletes that term:

\nabla^2\vec A=-\mu_0\vec J \quad\Longleftrightarrow\quad \nabla^2A_x=-\mu_0J_x,\;\; \nabla^2A_y=-\mu_0J_y,\;\; \nabla^2A_z=-\mu_0J_z . \qquad (4)

where: each line is literally electrostatics — $\nabla^2V=-\rho/\varepsilon_0$ with the cast changed. This is the promised reduction, the one the curl-of-B article previewed by name. Everything E1 and E2 built now transfers: the Green-kernel solution $\frac{\mu_0}{4\pi}\int\frac{J_i}{\mathfrak r}\mathrm{d}\tau'$ (which the previous article certified independently by direct curl — the two articles meet in the middle), uniqueness theorems, separation of variables, relaxation methods, the method of images. Magnetostatics has been reformatted as three electrostatics problems plus one curl.

Limiting cases

case	result	meaning
Coulomb gauge, localized currents, $\vec A\to0$	unique $\vec A=\frac{\mu_0}{4\pi}\int\frac{\vec J}{\mathfrak r}\mathrm{d}\tau'$	the standard representative; previous article's formula
Coulomb gauge, unbounded field region	representative unique only up to harmonic $\nabla\mu$	the uniform-field menagerie's coexistence
Lorenz gauge $\nabla\cdot\vec A=-\mu_0\varepsilon_0\,\partial_tV$	wave equations for $V,\vec A$	the relativistic workhorse — named here, owned by the radiation chapter
statics	Lorenz and Coulomb coincide	$\partial_t=0$ erases their difference; no decision needed yet
multiply connected region, same circulations	gauge function single-valued anyway	topology fine print of Derivation A
gauge-dependent quantity in a final answer	a bug, by definition	the gauge-invariance audit

Common trapFixing a gauge changes nothing physical — so never let a gauge choice into a physical answer

Two failure modes, mirror images. The first is fear: treating gauge fixing as an approximation, hedging results with "in the Coulomb gauge" as if $\vec B$ , forces, or fluxes might differ elsewhere. They cannot — gauge choice is map projection, and every invariant comes out identical in every gauge; choose the convenient one with a clear conscience. The second is carelessness: letting gauge-dependent intermediate quantities leak into final answers. The value of $\vec A$ at a point, the location of its zeros, the split $\vec p=m\vec v+q\vec A$ — none is measurable, and a result that changes under $\vec A\to\vec A+\nabla\lambda$ is wrong before it is checked against experiment. Related trap, inherited from equation (3): the componentwise Poisson reduction (4) is a Cartesian-components statement; writing $\nabla^2A_\phi=-\mu_0J_\phi$ in cylindrical coordinates is false (the vector Laplacian couples $A_s$ and $A_\phi$ ), a mistake that produces plausible-looking nonsense in solenoid problems.

Worked examples

1 · Two uniforms for one field (confidence builder)

Take $B=0.50$ T along $\hat z$ . Verify that the Landau potential $\vec A_L=-By\,\hat x$ and the symmetric potential $\vec A_S=\tfrac12 B(-y\,\hat x+x\,\hat y)$ describe it equally, find the gauge function connecting them, and evaluate everything at the point $(x,y)=(2.0\ \mathrm{cm},\,3.0\ \mathrm{cm})$ .

Step 1. Same curl. $\nabla\times\vec A_L=-\partial_y(-By)\,\hat z=B\hat z$ ✓ and $\nabla\times\vec A_S=\tfrac12B\,\hat z+\tfrac12B\,\hat z=B\hat z$ ✓. Both are even Coulomb: $\nabla\cdot\vec A_L=0$ , $\nabla\cdot\vec A_S=0$ .

Step 2. The connecting λ. $\vec A_S-\vec A_L=\tfrac12B(y\,\hat x+x\,\hat y)=\nabla\!\left(\tfrac12Bxy\right)$ , so $\lambda=\tfrac12Bxy$ — harmonic ( $\nabla^2\lambda=0$ ), as Step 3 of the fixing derivation requires for a Coulomb-to-Coulomb move.

Step 3. Numbers at the point. $\vec A_L=(-0.015,\,0,\,0)$ T·m; $\vec A_S=(-0.0075,\,+0.0050,\,0)$ T·m, magnitude $9.0\times10^{-3}$ T·m; $\lambda=\tfrac12\times0.5\times0.02\times0.03=1.5\times10^{-4}$ T·m². Different vectors, different magnitudes, different directions — one field.

Step 4. The invariant audit. Circulation around the square with corners $(0,0),(L,0),(L,L),(0,L)$ , $L=2$ cm: Landau — only the top leg contributes, $(-BL)(-L)=BL^2$ ; symmetric — the legs along $x=L$ and $y=L$ contribute $\tfrac12BL^2$ each while the two legs through the origin contribute nothing, total $BL^2$ . Both give $\Phi=0.5\times(0.02)^2=2.0\times10^{-4}$ Wb ✓. The maps disagree pointwise and agree on every loop — gauge freedom in one worked page.

2 · Repairing a potential into the Coulomb gauge (exam level)

The potential $\vec A_0=\left(\tfrac{-By}{2}+2xy\right)\hat x+\left(\tfrac{Bx}{2}+x^2\right)\hat y$ also describes $B\hat z$ (check its curl). It is not Coulomb. Repair it.

Step 1. Diagnose. $\nabla\cdot\vec A_0=\partial_x\!\left(\tfrac{-By}{2}+2xy\right)+\partial_y\!\left(\tfrac{Bx}{2}+x^2\right)=2y+0=2y$ . (The curl: $\partial_x(\tfrac{Bx}{2}+x^2)-\partial_y(\tfrac{-By}{2}+2xy)=\tfrac B2+2x+\tfrac B2-2x=B$ ✓ — same field, wrong gauge.)

Step 2. Solve the repair equation. Need $\nabla^2\lambda=-2y$ . Try a pure power: $\lambda=-y^3/3$ gives $\nabla^2\lambda=-2y$ ✓. (Any solution works; others differ by harmonic functions — that is Step 3 freedom, not error.)

Step 3. Transform. $\vec A=\vec A_0+\nabla\lambda=\vec A_0+(0,-y^2,0)$ :

\vec A=\left(\tfrac{-By}{2}+2xy\right)\hat x+\left(\tfrac{Bx}{2}+x^2-y^2\right)\hat y .

Audit: $\nabla\cdot\vec A=2y-2y=0$ ✓, $\nabla\times\vec A=B\hat z$ ✓ (recomputed symbolically).

Step 4. Read the leftover. The repaired $\vec A$ is not the symmetric gauge — it exceeds it by $(2xy,\,x^2-y^2,\,0)=\nabla(x^2y-\tfrac{y^3}{3})$ , and $\nabla^2(x^2y-\tfrac{y^3}{3})=2y-2y=0$ : a harmonic gauge function, exactly the residual freedom the theory predicted. To land on the symmetric representative specifically, you must impose more than $\nabla\cdot\vec A=0$ — a boundary/growth condition. Gauge fixing is a two-lock door: the differential condition is the first key, boundary behavior the second.

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §5.4, §10.1.2 · Jackson, Classical Electrodynamics, 3rd ed., §5.4, §6.2–6.3 · Zangwill, Modern Electrodynamics, §10.5 · Jackson & Okun, "Historical roots of gauge invariance," Rev. Mod. Phys. 73, 663 (2001) The key vector identities underlying this article are $\nabla\times(\nabla\lambda)=0$ and $\nabla\cdot(\nabla\times\mathbf{A})=0$ ; the curl-of-curl identity reduces Ampère's law to $\nabla^2\mathbf{A}=-\mu_0\mathbf{J}$ in Coulomb gauge. The permeability of free space takes the value $\mu_0=4\pi\times10^{-7}\,\mathrm{T\,m/A}$ (the historical defining value; post-2019 SI fixes it via measurement, with the value essentially unchanged). The integral solution is derived in the previous article by direct curl; this article derives the Poisson route and the gauge theory around it. Symbolic checks for the worked examples: both curls equal $B\hat z$ ; $\lambda=\tfrac12Bxy$ is harmonic; the repair gauge function is $\lambda=-y^3/3$ ; the residual $x^2y-y^3/3$ is harmonic.

Builds on: Magnetic Vector Potential · Curl of B · Ampere's Law · Poisson's Equation · Uniqueness Theorems

Leads to: Multipole Expansion of the Vector Potential · Magnetic Dipole and Magnetic Moment · Boundary Conditions for the Magnetostatic Field

Is gauge freedom telling me A has no physical meaning at all?

No — it is telling you precisely which part is physical. Think of the information in $\vec A$ as circulation data: every closed loop's $\oint\vec A\cdot\mathrm{d}\vec\ell$ is gauge-invariant, and the full set of them (equivalently $\vec B$ plus the fluxes through any holes in the region) is exactly what all gauges agree on and exactly what experiments can access — including quantum ones, per Aharonov–Bohm. The pointwise values are the non-physical packaging around that content, like a mountain's altitude before you fix sea level. The previous article's slogan survives sharpened: real field, redundant description; gauge theory is the discipline of never confusing the two.

Why pick div A = 0 rather than some other condition?

Because of what it deletes. Ampère's law in potential form carries the term $\nabla(\nabla\cdot\vec A)$ , which couples the three components of $\vec A$ into one tangled system; setting $\nabla\cdot\vec A=0$ removes it surgically, leaving three independent, boundary-tested Poisson equations. Other conditions are better for other jobs: the Lorenz gauge treats space and time symmetrically and turns the full Maxwell equations into wave equations (the radiation chapter's tool); the Landau gauge makes a uniform field's quantum problem separable; axial gauges kill one component outright. In statics the Lorenz and Coulomb conditions coincide, so nothing is even given up. "Which gauge?" is always "which calculation?" — never "which physics?".

If lambda is arbitrary, why does the gauge function in the repair problem come out specific?

Two different λ's are in play, and the distinction is the concept. The transformation λ is arbitrary — you may apply any of them, that is the freedom. The repair λ is the particular one that lands you on a chosen target condition, and solving for it is what "gauge fixing" means operationally: demand $\nabla\cdot(\vec A_0+\nabla\lambda)=0$ and λ must obey $\nabla^2\lambda=-\nabla\cdot\vec A_0$ — arbitrary no longer, though still not unique (any harmonic function may be added, the residual freedom). Compare choosing a map projection: you may draw the map any way at all, but once you announce "Mercator," the drawing instructions are fixed up to trivial leftovers.

The three uniform-field potentials look completely different. Would a charged particle really move identically in all three?

Identically — trajectories, periods, radii, everything observable. The Lorentz force is $q\vec v\times\vec B$ and all three give the same $\vec B$ . What differs is the bookkeeping of the motion: the canonical momentum $\vec p=m\vec v+q\vec A$ is gauge-dependent, so which components of $\vec p$ are conserved differs — Landau gauge conserves $p_x$ (its $\vec A$ ignores $x$ ), the symmetric gauge conserves canonical angular momentum (its $\vec A$ ignores the azimuth). Same circles, different conserved labels for them. In quantum mechanics the same story: the Landau-gauge eigenfunctions are plane-wave strips, the symmetric-gauge ones are angular-momentum rings, the energy spectrum (Landau levels) is identical, and unitary gauge transformation maps one basis onto the other.

What breaks in a region with a hole, like outside a solenoid?

The converse theorem's construction. A curl-free difference field $\vec W$ defines $\lambda$ by path integration, and path independence needs every loop to be shrinkable; around the solenoid's hole a loop cannot shrink, and $\lambda$ can return from the round trip incremented — multivalued. For gauge functions physics closes the loophole: two potentials describing the same field must agree on the hole's circulation (it is the gauge-invariant flux), so their difference has none and its λ is single-valued. But the same mathematics with the roles shifted explains a genuine phenomenon: $\vec B$ itself is curl-free outside the wire yet has circulation $\mu_0I$ around it, so its would-be scalar potential is irreparably multivalued — which is why magnetostatics needed a vector potential in the first place, and why the magnetic scalar potential (used for convenience in the magnetized-matter chapter) only lives in current-free, hole-managed regions.

Does fixing the gauge lose solutions or physics?

Neither, and the theorem pair of this article is the warranty. Completeness: every physical field configuration has at least one Coulomb-gauge description, because the repair equation $\nabla^2\lambda=-\nabla\cdot\vec A_0$ always has a solution (existence, Step 2). Faithfulness: gauge transformations never change $\vec B$ or any invariant, so restricting attention to one representative discards only duplicates, never physics (the converse theorem says all the duplicates are gradient-related — nothing else lurks). The one genuine subtlety is bookkeeping at the margins: with nontrivial topology or fields that do not decay, the residual harmonic freedom means "the" representative needs boundary data to pin down — carelessness there produces apparent paradoxes, not lost physics.

Is the Coulomb gauge still available when fields depend on time — and is it still the best choice?

Available, always: the repair argument never used statics. But its character changes and its convenience usually evaporates. In time-dependent theory the Coulomb-gauge scalar potential obeys $\nabla^2V=-\rho/\varepsilon_0$ instantaneously — the potential updates everywhere the moment charge moves, with no light-travel delay. That looks like a causality scandal; it is not, because $V$ alone is gauge-dependent and unobservable, and the physical fields $\vec E=-\nabla V-\partial_t\vec A$ conspire to propagate strictly causally — a beautiful consistency check worked out in the radiation chapter. For genuinely dynamical problems the Lorenz gauge, which makes both potentials obey causal wave equations sourced by retarded integrals, is almost always the cleaner instrument. In this chapter, statics makes the two coincide, and no decision is required.